[rudi_mk]

Disclaimer - been working on K8s in different capacities since 2017, so there’s definitely a lot of bias.

My take - Kubernetes isn’t perfect. It’s got a lot of moving parts, a lot of complexity. My work involves managing SRE for almost a hundred clusters daily, where clusters are running across different clouds with different levels of customisation. I’ll say this - K8s is to cloud native infrastructure engineering what ODBC was for databases 20 years ago. A consistent base standard for running workloads across public, private and hybrid clouds, without worrying _too much_ about how those apps will actually utilise the underlying infrastructure. I’ve worked on Docker Swarm, Rancher 1.0, Mesos, Nomad - nothing comes close.

Sure things may evolve and maybe in a decade we _may_ see a new standard, perhaps a new platform. But I don’t really see that happening. K8s is here to stay, and it’s now at a point where it’s stable and mature from a reliability standpoint; it just needs people like us to worry and fret about making it more accessible.

[scombridae]

My take - Kubernetes isn’t perfect

Good to know. When I ask if a technology is worth leanring, I always assume it's perfect. I was also relieved to discover a new platform might but also might not surface to overtake K8s.

[rudi_mk]

Keeping the complexity aside - which is always a decent motivator for new platforms - I feel a significant amount of technical and financial investment has been made in Kubernetes over the last few years. So when a community like the CNCF comes together and says okay, where do we go from here, I don’t think anybody says let’s ditch all we’ve done so far for a new platform that may never happen. Their focus ends up trying to either extend what we have already - which, let’s be fair, does actually work - or they look for ways to make using Kubernetes slightly easier.

[leoh]

You’re conflating connotation with denotation ;-)

Saying “it isn’t perfect” usually means something has fairly fundamental flaws that make it uncomfortable to use; and can simultaneously imply that it’s among the best we’ve got.

Almost certainly something will replace it or make it far more reliable, understandable, and ergonomic over time.

[scombridae]

imply that it’s among the best we’ve got.

If this and GP's response are inception-level sarcasm, then I'm outclassed.

My initial reaction to GP was intended to cast a light on its vacuousness, what with all the hedging and credentialling.

[rudi_mk]

I’m sorry you found my earlier response vacuous. The point I was trying to make is that K8s is the best shot we have, and I wanted to underline the possible bias I may have. I’m sure someone must have found what I said worth their time, just as someone won’t.

[antman]

Any go-to k8 tools that you consider important for most deployments?

[rudi_mk]

Not a big fan of tools like kops or even eksctl - sure they’re good for creating cookie cutter clusters, but I prefer using Terraform for cluster orchestration. Apart from that, k9s is a terrific tool for managing clusters instead of kubectl. Apart from that - a Grafana installation locally on my Mac with preloaded dashboards, and all I need to do is port forward to Prometheus, in order to view monitoring data easily.